Inspiration
Spatial scheduling of blocks is a real, costly bottleneck in shipyards: large blocks must be placed in limited space and scheduled under time and precedence constraints. We wanted an approach that performs well across all kinds of instances, not just the easy ones.
What it does
It solves the Grand Shipyard Puzzle (packing + scheduling) by finding an efficient assignment of blocks across space and time, always returning a valid solution within the time budget.
How we built it
A portfolio of complementary solvers (a fast constructive heuristic, an exact core on subproblems, and a local-search refiner) runs within the time budget, and a selector keeps the best result for each instance. The design is anytime — it always has a valid solution ready — and modular, so new strategies plug in without rewrites.
Challenges we ran into
Coupling packing and scheduling without the model blowing up, and keeping high-quality solutions within the time limit ("Beat the Clock") on the largest instances.
Accomplishments that we're proud of
Settling on a robust, modular architecture and a plan that gets us onto the leaderboard from day one, with a clear path to keep improving instead of betting everything on a single fragile method.
What we learned
That in these competitions robustness tends to beat fragile brilliance, and that a fast local evaluation setup — letting us test our own solutions without waiting on the server — is the biggest driver of fast improvement.
What's next for FlowPack — Adaptive Portfolio Block Optimizer
Building the evaluation harness and a baseline-beating submission, then tuning the per-instance selector and adding stronger refinement operators throughout the preliminary round.
Log in or sign up for Devpost to join the conversation.